8:42 AM - Inbox triage
I open the dashboard. Three new SMS alerts from overnight. The Slack notification badge is blue, which means nothing urgent from the agent logs. I click through to the inbox.
Monday night was busy. Twenty-three calls came in after 6 PM across all my customers. The agent flagged three for me to review manually. I scroll through the queue. Most calls are booked cleanly: a water heater replacement at Brennan's Plumbing scheduled for Thursday, two HVAC callbacks at Rodriguez Climate Control for today and Wednesday, a consultation request for estate planning at Williamson Law that's now queued for the first available opening this week. I approve each one with a checkbox. The agent learns what a good intake looks like.
The fourth alert is orange. I click it. A caller at Carol Reyes's clinic called asking about migraine treatment covered under specific insurance plans. The agent captured the question but couldn't determine urgency from the caller's tone alone. No chest pain, no acute symptoms, but the caller sounded worried. I read the transcript. The agent offered a callback within two hours and the caller accepted. That's the right move. I mark it handled.
I open my email. Slack has been quiet, which is good. But there's an email from Steve Chen at Brennan's Plumbing in my inbox sent at 6:47 AM. Subject: "Couple things." I open it.
The first paragraph: the agent booked a call on Thursday at 3 PM, but Steve's team only works until 2 PM on Thursdays. He's asking if the agent can call the customer and move it or if he has to do it. The second paragraph is easier. He's thanking me for catching a lead last week that turned into a $3200 water heater job and says he's thinking about signing up for the premium tier next month.
I note the first issue. The agent's instructions say 2 PM upper limit for Brennan's on Thursdays. I added that limit three weeks ago, but apparently it didn't take. I'll check the instructions in the admin panel and adjust. For now, I send Steve a quick Gmail reply letting him know I'll have the agent call and reschedule this morning. That's on me.
9:55 AM - The metrics check
I pull up the analytics tab. Tuesday is always slower than Monday, but the pipeline number makes me sit back. Week-to-date, I've booked 47 appointments across my five customers. Seven of those are signed proposals (shown as converted leads in the agent's output). That's a $840 weekly base revenue, assuming my customers close at the 15% booking-to-signed rate I've been tracking. Not enough to live on yet, but I signed my seventh customer yesterday, so this number will accelerate in the coming weeks.
Today's numbers are fresh. Four new calls came in before 9 AM. Three are now on the calendar. One is a pricing inquiry from a new potential customer asking what our monthly fee would be if Williamson Law brought us on. I make a note to follow up on that cold lead email thread myself.
I open my Stripe dashboard in another tab. Last month, AfterHours brought in $1,847 from active customer subscriptions. Three of those are on the $149 tier, two on $199, one on $249 (premium, with higher call volume and CRM integration). This month, I'm on track for $2,100 if the two new customers stay through month-end. Not enough, but the trajectory is there. I can see the edges of something real.
10:32 AM - The rescheduling call
I navigate to Brennan's Plumbing's configuration in the admin panel. I check the instructions. There it is: "No appointments after 2 PM Thursday." But the agent scheduled at 3 PM. I look at the revision history. The instructions were updated on April 18, but there's a note in the logs that the agent's context was cached from April 15. The system didn't refresh the updated rules. This is a known bug in the product documentation. I email the support channel with the case details and a request to force-refresh the agent's system prompt. While I wait for that, I pull the customer number from the Thursday 3 PM booking and dial.
The voicemail picks up. I leave a message identifying myself as the automated assistant's operator, apologize for the scheduling conflict, and offer two new times: 1:30 PM Thursday or 10 AM Friday. I mention I'm calling personally to make sure they get what works best. I hang up and log the manual intervention in the Slack #escalations channel, tagging it as a system-level bug that needs fixing.
This kind of thing happens once or twice a week. It's the work that doesn't ship in a blog post about AI, but it's the work that keeps customers happy.
12:15 PM - Lunch and the churn alert
I grab coffee and open my email again. There's a message from Maria at Rodriguez Climate Control from forty minutes ago. Her subject line is "Thinking of canceling." I read it cold. She says three out of the last five calls the agent handled were for jobs Rodriguez Climate doesn't actually do. The caller was asking about furnace repair, but Rodriguez only does HVAC installation and maintenance. The agent offered to schedule anyway, and now Maria has three callbacks she doesn't want, taking up her morning tomorrow.
This is the second complaint from Maria in two weeks. The first was about the agent not catching that a caller had already called once before. That one was my fault - the agent didn't have customer call history integrated into its context. I promised to fix it. I didn't.
I open a blank Gmail draft. "Hi Maria, Thanks for the feedback. You're right on both counts. I should have mapped your service boundaries in the agent's system before you went live. Let me do that today. I'm going to go through the last three calls and send you suggested replies you can use, and I'm adding a check: no callbacks unless the job matches your service list. One question: would premium-tier CRM integration help you see incoming calls in real time so you can jump in if the agent gets it wrong? Let me knock 20% off next month while we sort this out." I hit send. I don't know if she'll stay, but ghosting her would definitely lose her.
I make a note to myself: This is the difference between a product that works in a demo and a product that works in a customer's actual business. Integration depth matters. I spend the next 30 minutes updating Rodriguez's agent profile with a hard list of service categories and a note in the system prompt about the call history integration that needs to happen.
2:48 PM - A customer thank you
My phone buzzes. Text message from Carol Reyes at Carol Reyes's Clinic. "The patient we were worried about this morning? Called back. Turns out she's fine, just medication-related. But she said the message the automated voice left was professional and calm, which made her feel taken care of even when the office was closed. Really good experience. Billing question though - can I pay quarterly instead of monthly? Moving to your premium tier."
I smile and text back: "Absolutely. I'll set up a quarterly option. Great news about the patient. Call in 10 min?" We talk through the integration details. She wants to see inbound calls logged to her EHR system so there's a paper trail. That's a feature request. I add it to the linear board I'm tracking in Linear and mark it as customer-requested. It goes on the backlog.
I update her subscription in Stripe. Quarterly at the premium rate adds $597 to the projected monthly recurring revenue once it processes. Still not enough to hire anyone, but one more week of new customers at this rate and I can drop my freelance translation work and do this full-time.
4:33 PM - Pipeline and a small fix
I open the admin dashboard again. This week I'm tracking which call types convert fastest. Appointments booked same-day have a 68% show-up rate. Appointments booked more than a week out drop to 41%. That's going to change how I brief the agent on urgency language. I add a note to make the agent more aggressive about same-day booking when the customer seems flexible.
I also notice one more alert in the logs. Williamson Law's agent tried to book an appointment during the attorney's lunch hour (12-1 PM, which is blocked on the calendar). The agent caught the conflict and offered two adjacent times instead. The customer chose 11:15 AM. Clean execution. I approve it and make a mental note to show the example in the next customer onboarding call.
One small bug from the morning is still bothering me. I dive into the admin panel and find the cached context issue. I can't force a refresh from the operator dashboard, so I file a more detailed ticket in the product's support channel with reproduction steps. The issue is noted by support as a known limitation in the current version. I make a note that the next product update will include a manual refresh button. That's a priority fix for reliability.
5:54 PM - Evening close
I open my notebook. Tuesday is closing out. Let me count the actual output:
- 23 calls handled overnight, 21 booked or captured cleanly
- One manual rescheduling intervention due to a system bug (45 minutes, including diagnosis)
- One customer email drafted and sent in response to churn risk (30 minutes)
- One customer thank-you and upsell completed (20 minutes)
- Three hours of integration work and customer profile tuning
- One product bug escalated
Revenue today: $3 from a new monthly subscription (pro-rated). Pipeline: 11 appointments booked today, all on track to close or become callbacks. Week-to-date runway is looking like $2,200 in recurring revenue by month-end, up from $1,847 last month.
The work that felt real today: hearing Carol say the experience made her patient feel cared for. That's not a bot. That's a tool I configured to amplify my ability to show up for customers even when I'm asleep. The work that felt hard: knowing I let Maria down on system setup, and having to rebuild her trust by doing the work myself.
I close the laptop around 6:10 PM. This isn't passive income. It's not a script that runs while I sleep and deposits money. It's a business where I'm still the operator, the one who has to diagnose bugs, handle edge cases, and listen to customers when things don't work. The agent is the tool that keeps me from drowning, not the tool that replaces me. And if I keep showing up the way I did today - catching problems early, fixing what breaks, asking customers what they actually need - this thing could grow into something real.